Kwai-Keye/Thyme-RL
Kwai-Keye/Thyme-RL is a 7 billion parameter multimodal model with a 32768-token context length, developed by Kwai-Keye. It is designed to "Think Beyond Images" by autonomously generating and executing image processing and computational operations through executable code. This model excels at high-resolution perception and complex reasoning tasks, leveraging a novel two-stage training strategy combining supervised fine-tuning with reinforcement learning and the GRPO-ATS algorithm.
Loading preview...
Thyme-RL: Think Beyond Images
Thyme-RL, developed by Kwai-Keye, is a 7 billion parameter multimodal model with a 32768-token context length that redefines image understanding by moving beyond traditional "thinking with images" paradigms. It achieves this by autonomously generating and executing diverse image processing and computational operations through executable code. This capability significantly enhances its performance on high-resolution perception and complex reasoning tasks.
Key Capabilities
- Autonomous Code Execution: Generates and executes image processing and computational code.
- Enhanced Perception: Excels at high-resolution image perception.
- Complex Reasoning: Demonstrates strong performance in complex reasoning tasks.
- Novel Training Strategy: Utilizes a two-stage training approach combining supervised fine-tuning (SFT) with reinforcement learning (RL).
- GRPO-ATS Algorithm: Leverages the innovative GRPO-ATS algorithm for a sophisticated balance between reasoning exploration and code execution precision.
Good for
- Applications requiring advanced image understanding and manipulation.
- Tasks involving complex visual reasoning and problem-solving.
- Scenarios where autonomous execution of image processing operations is beneficial.
- Research into multimodal AI and reinforcement learning for visual tasks.